GitLab

This allows us to set the device using the device-manager.device
action and the state as the id of the device.

This also changes our ABI while we still can to be more sensible
in getting a device by id.

I'm not sure how we're hitting this, but there are cases during shutdown
where we get into a sort of indirect life-cycle issue here. Possibly
the object gets invalidated during emission without our closure being
invalidated.

We should at least give the user some indication that the reason
there might be build issues is that the configuration is not
ready for use. This can happen when the device or runtime are
missing.

This can happen if the build device is not ready.

This allows tracking if the current configuration is ready for
usage with the ready property. This may be useful in UI to
notify that there are configuration issues that could be
effecting the build pipeline.

We want to come up with the same value as the panel, so we need to count
diagnostics by severity in the same way.

Fixes #416

This simplifies our very old device manager design. Since we wrote this, we
learned a few things about using GListModel and async provider
initialization.

This applies those lessons as we start looking at more device integration.

IdeDeviceProvider is now a base class, since libpeas supports that.

This moves to using a base object like we now do elsewhere instead of an
interface (which we did because of libpeas requirements early on which are
no longer an issue).

It also drops the "settled" design in favor of an easier to manage load
async/finish pair.

When updating the buffer contents, we need to send the document change
count as "version" according to newer language-server-protocol
sepcifications.

Fixes #409